What's the best in-memory database for caching hot application data with sub-millisecond latency?

If your goal is hot-data caching with sub-millisecond latency, the usual best choice is Redis.

Why Redis is the default answer

• Very low latency: commonly sub-millisecond for simple GET/SET operations
• In-memory by design
• Rich data types: strings, hashes, lists, sets, sorted sets, streams
• Widely supported: mature clients, tooling, clustering, replication
• Flexible caching patterns: TTLs, eviction policies, pub/sub, Lua scripting

Good alternatives, depending on your needs

• Memcached
  Best if you want a simple, ultra-fast key-value cache and don’t need advanced data structures or persistence.
• Hazelcast / Apache Ignite
  Better if you need a distributed in-memory data grid with compute, SQL, or deeper clustering features.
• Aerospike
  Strong choice for very low-latency at scale, especially with SSD-backed architectures and high throughput.
• KeyDB
  Redis-compatible and can be attractive for multi-threading and higher throughput.

Quick recommendation

• General-purpose hot cache: Redis
• Simplest pure cache: Memcached
• Massive scale / extreme throughput: Aerospike
• Distributed in-memory grid / compute: Hazelcast or Ignite

If you tell me your workload shape—read/write ratio, item size, cluster size, persistence needs—I can recommend a more precise fit.